Assessing Out-of-Domain Language Model Performance from Few Examples

TL;DR

This paper investigates predicting out-of-domain performance of language models using few-shot examples and attribution analysis, finding that simple accuracy measures are surprisingly effective but attribution factors can provide additional insights.

Contribution

It introduces a method to forecast out-of-domain performance of language models with few-shot data, incorporating feature attribution analysis to identify potential generalization issues.

Findings

Accuracy on few-shot examples is a strong baseline.

Attribution-based factors can help rank models' out-of-domain performance.

Simple accuracy measures often outperform complex attribution analysis.

Abstract

While pretrained language models have exhibited impressive generalization capabilities, they still behave unpredictably under certain domain shifts. In particular, a model may learn a reasoning process on in-domain training data that does not hold for out-of-domain test data. We address the task of predicting out-of-domain (OOD) performance in a few-shot fashion: given a few target-domain examples and a set of models with similar training performance, can we understand how these models will perform on OOD test data? We benchmark the performance on this task when looking at model accuracy on the few-shot examples, then investigate how to incorporate analysis of the models' behavior using feature attributions to better tackle this problem. Specifically, we explore a set of "factors" designed to reveal model agreement with certain pathological heuristics that may indicate worse generalization capabilities. On textual entailment, paraphrase recognition, and a synthetic classification task, we show that attribution-based factors can help rank relative model OOD performance. However, accuracy on a few-shot test set is a surprisingly strong baseline, particularly when the system designer does not have in-depth prior knowledge about the domain shift.